Contrast Affects Flicker and Speed Perception Differently

Transcription

1 Pergamon PH: S (96)32.1 Vision Res., VoL 37, No. 1, pp , Elsevier Science Ltd. All rights reserved Printed in Great Britain $17. + : Contrast Affects Flicker and Speed Perception Differently PETER THOMPSON, *:j: LELAND S. STONEt Received 7 Agst 1996; in revised form 8 November 1996 We have previosly shown that contrast affects speed perception, with lower-contrast, drifting gratings perceived as moving slower. In a recent stdy, we examined the implications of this reslt on models of speed perception that se the amplitde of the response oflinear spatio-temporal filters to determine speed. In this stdy, we investigate whether the contrast dependence of speed can be nderstood within the context of models in which speed estimation is made sing the temporal freqency of the response of linear spatio-temporal filters. We measred the effect of contrast on flicker perception and fond that contrast maniplations prodce opposite effects on perceived drift rate and perceived flicker rate, i.e., redcing contrast increases the apparent temporal freqency of conterphase modlated gratings. This finding arges that, if a temporal freqencybased algorithm nderlies speed perception, either flicker and speed perception mst not be based on the otpt of the same mechanism or contra!>t effects on perceived spatial freqency reconcile the disparate effects observed for perceived temporal freqency and speed Elsevier Science Ltd. All rights reserved. Motion perception Speed discrimination Temporal freqency Conterphase flicker Contrast Area VI Area MT INTRODUCTION There is considerable psychophysical (e.g. Anderson & Brr, 1985; Anderson et al., 1991; Watson & Trano, 1995) and physiological (e.g. Movshon et al., 1978; Hamilton et al., 1989) evidence that the first stage of visal cortex processing decomposes the image into its spatio-temporal freqency components. This fact inspired the development of models of hman motion processing that se the otpt of directionally selective linear spatio-temporal filters as inpt (Watson & Ahmada, 1983, 1985; Adelson & Bergen, 1985). Becase the otpt of visal cortical nerones (Dean, 1981; Albrecht & Hamilton, 1982; Sclar et ai., 199) and their theoretical idealizations (Watson & Ahmada, 1983; Albrecht & Geisler, 1991) depend on contrast as well as speed, modellers of hman motion perception were prompted to develop varios schemes to overcome this problem and to generate velocity estimates robst to changes in contrast (Watson & Ahmada, 1985; Adelson & Bergen, 1986). In particlar, Adelson & Bergen (1986) proposed a model in which speed is derived by dividing (normalizing) the otpt of motion-energy nits by a 'Department of Psychology, University of York, York, U.K. tflight Management and Hman Factors Division, NASA Ames Research Center, Moffett Field, CA , U.S.A. tto whom all correspondence shold be addressed [Fax: ; pt2@york.ac.kj term related to the average contrast, thereby prodcing accrate speed estimates independent of contrast (except possibly at very low contrast). However, sch mechanisms are seemingly at odds with or previos finding that hman speed estimates do indeed depend on contrast over a wide range of contrasts (Thompson, 1976, 1982; Stone & Thompson, 1992). In an effort to reconcile these reslts with motion-energy models, Stone & Thompson (1992) sggested that a contrast-normalization scheme might nonetheless explain the observed speed misperception, if the spatio-temporal window over which the normalization occrred were large. In this case, each patch might interfere with the other's normalization and prodce the observed contrast dependence of the speed matches. However, recent experiments have rled ot this version of normalization as well (Thompson et al_, 1996), leading s to search for other possible explanations for the phenomenon. A different approach for achieving contrast-robst speed estimates was devised by Watson & Ahmada (1985). Their model comptes speed from the temporal freqency of the response of the linear spatio-temporal filters, becase sch neral signals are largely independent of contrast (e.g. Albrecht & Geisler, 1991, 1994). If speed is compted by sing the temporal freqency of the response of spatially tned channels, then it seems reasonable to expect that any effect of contrast on the perceived speed of a drifting grating shold also be 1255

2 1256 P. THOMPSON and L. S. STONE A deg 14 deg 2 deg B e -, 5 ms 5 ms 5 ms L FIGURE 1. Stimls configration for (A) Experiment 1 in which perifoveal stimli were presented and (B) Experiment 2 in which foveal stimli were presented. The temporal seqence of events in both experiments is shown in (C). In both experiments the windowing of the stimli was sharp in both space and time. manifest in the perceived temporal freqency of a conterphase flickering stationary grating. In this stdy, we examine the effect of contrast on speed and temporal freqency nder identical experimental conditions. We find that both are indeed affected by contrast bt that the effects are opposite in direction. A preliminary report of or findings was made at the annal meeting of the Association for Research in Vision and Ophthalmology (Thompson & Stone, 1996). "Becase or presentation intervals lasted 5 msec it is possible that or drifting stimli elicited smooth eye-movement responses despite the presence of a fixation point. However, given that or previos stdy docmenting the effect of contrast on perceived speed (Stone & Thompson, 1992) was performed sing a simltaneos spatial forced-choice task, the main effect of contrast on perceived speed cannot merely be the reslt of an eye-movement artifact, becase any eye movement wold have affected the motion of the two patches identically. Nonetheless, becase the present stdy ses a seqential task, and as the magnitde and latency of oclar following may depend on contrast (Miles et al., 1986), it is possible that a differential eye-movement response in the foveal vs peri foveal conditions cold be responsible for the small differences observed (see Reslts). Flickering gratings are nlikely to elicit smooth eye movements, therefore the above caveat does not apply to or flicker reslts, the main finding of this paper. GENERAL METHODS We performed two experiments. Experiment 1 investigated the dependence of speed and flicker jdgements pon contrast for perifoveally presented stimli, while Experiment 2 investigated these effects for foveal stimli. In both experiments, a trial consisted of two stimls intervals. In each interval, a horizontal, 2 cycles/ deg sinsoidal grating patch (2 deg wide by 1 deg high) was presented for 5 msec separated by a blank period of 5 msec in which only the mean lminance (7 cd/ m 2 ) was present." In Experiment 1, one patch was centred 1 deg above and one centred 1 deg below the fixation point (in random order), while in Experiment 2 both patches were centred on the fixation point (see Fig. 1). Within each experiment, two separate conditions were rn to investigate two types of discrimination. In the drift condition, observers were shown grating patches drifting pwards within each interval and were asked to report which of the two intervals contained the patch which moved faster. In the flicker condition, observers were shown conterphase flickering grating patches within each interval and were asked to report which of the intervals contained the patch that flickered more rapidly. In both tasks, one of the pair of gratings, the "standard", was always modlated at 4 Hz, the modlation of the

3 CONTRAST AFFECTS FLICKER AND SPEED PERCEPTION DIFFERENTLY 1257 other, the "test", was determined by a staircase procedre (Findlay, 1978). Each staircase terminated after a total of 12 reversals (abot 3 trials). We define the match as the ratio of the test modlation freqency to that of the standard (expressed as a percentage) at the point of sbjective eqality (determined by taking the mean of the last eight reversals). In all conditions reported here, for test-standard pairs of grating contrasts were investigated within interleaved independent staircases. Two baseline conditions consisted of standard and test gratings of eqal contrast, at 1 or 7% contrast. Two mixed-contrast conditions were rn: one with standard 1% and test 7% contrast, the other with standard 7% and test 1% contrast. Veridical perception wold yield a mean match of 1%. In Experiment 1 (Fig. 2), ten naive observers and one of the athors (PT) participated. In Experiment 2 (Fig. 3), nine naive observers and one of the athors (PT) participated. Six of the naive observers (in addition to PT) participated in both experiments. All conditions were rn three times by each observer. Stimli were generated on a Barco Calibrator 7651 screen sing a Cambridge Research Systems VSG 2.1 graphics display card (1 Hz refresh rate) hosed in a Compaq Deskpro 386/2 compter. Observers sat 114 cm from the screen at which distance the screen sbtended 18 deg by 14 deg of visal angle. The gamma nonlinearity of the monitor was corrected sing a look-p table.,-. 15 I-< 125 ril '+-< '-" 1...c:.. rj) 75 5 A nn ,-. 15 I-< B 125 ril '+-< B El c: 75 I-< C 1 11 ril '+-< 1...c: a- Drift Flicker Contrast Ratio (db) RESULTS Experiment 1,' Perifoveal presentation The speed matches for the drifting gratings presented perifoveally are shown in Fig. 2(A) for all 11 observers. In line with previos findings (Thompson, 1982; Stone & Thompson, 1992; Hawken et al., 1994; Ledgeway & Smith, 1995; Gegenfrtner & Hawken, 1996; Thompson et at., 1996), we fond that at low contrast perceived rate of motion is decreased and at high contrast perceived rate of motion is increased. The mean size of the effect (i.e., slope of speed match vs log contrast ratio from linear regression with 4 points) is.68%/db with y-intercept 12.% (mean r2:.81). Althogh there is clearly considerable inter-sbject variability in the size of the effect (the largest effect is 1.35%/dB while the smallest is.18%/db), all observers perceived lower contrast gratings as drifting slower (i.e., had positive slopes). The flicker matches for the conterphase gratings are shown in Fig. 2(B) for the same 11 observers. The effects of contrast are now reversed: increasing contrast decreases perceived flicker rate. The mean size of the effect is -.75%/dB with y-intercept 98.4% (mean r2:.79). Again, althogh there is a considerable range of inter-sbject variability in the data (from to FIGURE 2. Speed and flicker matching in the perifovea. For clarity, the data for contrast ratio zero are represented by the average of the two eqal contrast conditions. All P vales were obtained from npaired two-tailed I-tests. (A) The mean speed matches across three rns of all II observers in response to drifting gratings at three different contrast ratios. All observers showed an increase in perceived speed at higher contrast. For five observers, the difference between the matches in the 7: 1 and 1:7 conditions was significant across the small nmber of rns (P < (J.5) with two more observers borderline significant (P < )8). The average difference was very significant across observers (P <.(1) as well as the average slope (P <. I). (B) The mean flicker matches across three rns of the same II observers in response to flickering gratings at three different contrast ratios. Nine showed a decrease in perceived temporal freqency at higher contrast. Again, for five observers, the difference between the matches in the 7: I n and 1:7 conditions was significant across the small nmber of rns (P <.5) with two more observers borderline significant (P <.(9). The average difference was very significant across observers (P <.1) as well as the average slope (P < n.o(8). (C) Mean speed and flicker matches, with the error bars being standard error across observers.

4 1258 P. THOMPSON and L. S. STONE 15 A ; 125 [f) 'I-< o 1.:i:l ::E r , B t:< [f) 125 'I-< o -1 o 1 2 phase gratings are shown in Fig. 3(B) for the same 1 observers. Eight observers perceived the lower contrast gratings as flickering faster. The mean size of the effect was -.52%/dB (range: to.9%/db) with y intercept 1.4% (mean,.2:.79). These reslts are qalitatively the same as those seen in the peri fovea, again the magnitde of the effect appears smaller in the fovea. Given that the effects of contrast on flicker and speed perception are in opposite directions and of approximately the same size, it is tempting to think that the two effects might be mirror images of one another. However, we fond no significant correlation between observers' slopes for speed and flicker matches (r2 =.6). DISCUSSION McKee et ai. (1986) fond evidence that hmans can perceive speed with high precision, even in the presence Q..g of random flctations in stimls contrast. This finding = led theoretical neroscientists to generate comptational 1 models of hman motion processing that attempt to -5 generate speed estimates ncontaminated by stimls contrast. Given that motion-energy filters are inherently ::E 75 sensitive to stimls contrast, Adelson, Bergen, and ] others achieved robstness to variation in contrast by. sing a contrast-normalization procedre (Adelson & 5+--_ _----r-----, Bergen, 1986; Heeger, 1987; Wilson et ai., 1992). Yet, the fact that speed perception is contrast dependent over a wide range of contrasts (Stone & Thompson, 1992) c demonstrates that, if sch a normalization procedre is sed, it is only partially effective. The effectiveness of 1 normalization might be redced if the spatio-temporal 11 t:< <Il J::: a-- Drift ::E Flicker Contrast Ratio (db) O.15%/dB), nine observers perceived lower contrast gratings as faster (i.e., had negative slopes). Experiment 2: Foveal presentation The speed matches for the foveally presented stimli are shown in Fig. 3(A) for 1 observers. Nine observers perceived the lower contrast gratings as slower. The mean size of the effect was O.59%!dB (range: -.8 to 1.26%!dB) with a y-intercept of 1.8% (mean r2:.73). The effect is qalitatively the same as in the perifovea, althogh the magnitde of the effect appears smaller in the fovea. The flicker matches for the foveally presented conterwindow over which the normalization contrast is calclated were large enogh to case the two patches to interfere with each other's normalization. However, contrast-indced speed misperceptions are resistant to maniplations of srronding contrast and are observed even for matches across presentations made p to 5 sec apart in time (Thompson et at., 1996). These data rle ot the hypothesis that normalization is a rather global FIGURE 3. Speed and flicker matching in the fovea. For clarity, the data for contrast ratio zero are represented by the average of the two eqal contrast conditions. All P vales were obtained from npaired two-tailed t-tests. (A) The mean speed matches across three rns of all 1 observers in response to drifting gratings at three different contrast ratios. Nine showed an increase in perceived speed at higher contrast. For six observers, the difference between the matches in the 7:1 and 1:7 conditions was significant across the small nmber of rns (P <.5). The average difference between the matches in the 7:1 and 1:7 conditions was very significant across sbjects (P <.1) as well as the average slope (P <.3). (B) The mean flicker matches across three rns of the same 1 observers in response to flickering gratings at three different contrast ratios. Nine showed a decrease in perceived speed at higher contrast. For three observers, the difference between the matches in the 7:1 and 1:7 conditions was significant across the small nmber of rns (P <.5). The average difference between the matches in the 7:1 and 1:7 conditions was very significant across sbjects (P <.1) as well as the average slope (P <.6). (C) Mean speed and flicker matches, with the error bars being standard error across observers.

5 CONTRAST AFFECTS FLICKER AND SPEED PERCEPTION DIFFERENTLY 1259 process and that or specific spatio-temporal arrangement cased each patch to interfere with the other's otherwise proper normalization. For a more local version of normalization to explain or reslts, it wold have to be fndamentally incomplete. The otpt of motion filters is generally idealized as a power fnction of contrast with exponent n. Fll normalization posits that this raw otpt is divided by a measre of average contrast taken to the same exponent (see, e.g_ Albrecht & Geisler, 1991). If the exponent of the normalizing denominator is less than n, the normalization wold only be partially effective. Therefore, partial normalization might indeed explain or previos reslts. However, to explain or present reslts, different normalization wold be reqired for motion and flicker mechanisms. In conclsion, flly contrast-normalized motion-energy models are inconsistent with the observed contrast effect on perceived speed and flicker, althogh partial normalization schemes, local in space and time, are possible and deserve frther examination. Watson & Ahmada (1985) sggested an alternative method of compting speed, independent of contrast. Their model derives speed from the ostensibly contrastinvariant temporal freqency responses of directionally selective spatio-temporally tned inpt nits_ The original version of the model does not take into accont the fact that hman speed estimates are contrast dependent. However, if temporal freqency estimates within hman cortex are themselves contrast dependent, then perhaps a modified version of their model cold explain or previos reslts. To investigate this possibility, we examined the effect of contrast on temporal freqency estimation by measring flicker perception_ If the temporal freqency signal sed to compte stimls speed in their model were in fact contrast dependent, then the model might be able to explain or previos reslts. We reasoned that any effect of contrast on speed shold also be manifest in jdgements of temporal freqency. However, rather than finding that the effects on speed and flicker perception were the same, we fond them to have opposite sign. Three possible conclsions can be made from these reslts. First, it cold be the case that the Watson Ahmada model does not accrately describe the means whereby hmans determine speed. Second, it cold be that the flicker rate that we measred in or experiments is not tapping the same nderlying temporal freqency mechanism sed for speed estimation in the Watson Ahmada model and that something like their algorithm spports hman speed perception. Little in the literatre wold contradict sch a claim; McKee and colleages (1986) clearly believe that speed discriminations are not derived from flicker-rate discriminations, a view sppolied by Pasternak (1987). Thirdly, it cold be the case that the Watson-Ahmada model is correct and that the discrepancy between or perceived flicker and perceived speed jdgements as a fnction of contrast can be reconciled by the fact that perceived spatial freqency is also inflenced by contrast. In spport of this possibility, Georgeson (198) has demonstrated that redcing contrast increases perceived spatial freqency. Frthermore, the amplitde of this effect may be large enogh to override the effect on temporal freqency. This effect therefore has the potential to reconcile the observed increase in perceived temporal freqency and decrease in perceived speed with redced contrast, thereby rescing temporal-freqency based speed models. Ftre measrements of perceived spatial freqency, temporal freqency, and speed as a fnction of contrast nder identical conditions will be needed to determine if the three data sets are qantitatively linked as predicted by this explanation. REFERENCES Adelson, E. H. & Bergen, J. R. (1985). Spatiotemporal energy models for the perception of motion. Jornal of the Optical Society of America A, 2, Adelson, E. H. & Bergen, J. R. (1986). The extraction of spatiotemporal energy in hman and machine vision (pp ). Charleston, Soth Carolina: Institte of Electrical and Electronic Engineers Compter Society. Albrecht, D. G. & Geisler, W. S. (1991). Motion selectivity and the contrast-response fnction of simple cells in the visal cortex. Visal Neroscience, 7, Albrecht, D. G. & Geisler, W. S. (1994). Visal cortex nerons in monkey and cat: contrast response nonlinearities and stimls selectivity. In Comptational vision based on nerobiology (p. 254). Pacific Grove, CA: SPIE. Albrecht, D. G. & Hamilton, D. B. (1982). Striate cortex of monkey and cat: contrast response fnction. Jornal of Nerophysiology, 48, Anderson, S. J. & Brr, D. C. (1985). Spatial and temporal selectivity of the hman motion detection system. Vision Research, 25, Anderson, S. J., Brr, D. C. & Morrone, M. C. (1991). Twodimensional spatial and temporal freqency selectivity of motionsensitive mechanisms in hman vision. Jornal of the Optical Society of America A, 8, Dean, A. F. (1981). The relationship between response amplitude and contrast for cat striate cortical nerones..tornai of Physiology, London, 318, Findlay, J. M. (1978). Estimates on probability fnctions: a more virlent PEST. Perception and Psychophysics, 23, Gegenfrtner, K. R. & Hawken, M. J. (1996). Perceived velocity of lminance, chromatic and non-forier stimli: inflence of contrast and temporal freqency. Vision Research, 36, Georgeson, M. A. (198). Spatial freqency analysis in early visal processing. Philosophical Transactions of the Royal Society B, 29, Hamilton, D. B., Albrecht, D. G. & Geisler, W. S. (1989). Visal cortical receptive fields in monkey and cat: spatial and temporal phase transfer fnction. Vision Research, 29, Hawken, M. J., Gegenfrtner, K. R. & Tang, G. (1994). Contrast dependence of color and lminance motion. Natre, 367, Heeger, D. J. (1987). Model for the extraction of image flow. Jornal of the Optical Society of America A, 4, Ledgeway, T. & Smith, A. T. (1995). The perceived speed of secondorder motion and its dependence on stimls contrast. Vision Research, 35, McKee. S. P., Silverman, G. & Nakayama, K. (1986). Precise velocity discrimination despite random variations in temporal freqency and contrast. Vision Research, 26, Miles, F. A., Kawano, K. & Optican, L. M. (1986). Short-latency oclar following responses of monkey. I: Dependence on temporospatial properties of visal inpt. Jornal of Nerophysiology, 56, Movshon,1. A., Thompson, I. D. & Tolhrst, D. J. (1978). Spatial and

6 126 P. THOMPSON and L. S. STONE temporal contrast sensitivity of ncrones in areas 17 and 18 of the cat's visal cortex. Jornal of Physiology, London, 283, Pasternak, T. (1987). Discrimination of differences in speed and flicker rate depends on directionally selective mechanisms. Vision Research, 27, Sclar, G., Mansell, J. H. R. & Lennie, P. (199). Coding of image contrast in central visal pathways of the macaqe monkey. Vision Research, 3, Stone, L. S. & Thompson, P. (1992). Hman speed perception is contra:»t dependent. Vision Research, 32, Thompson, P. (1976). Velocity aftereffects and the perception of movement. Unpblished Doctoral Dissertation, University of Cambridge: Cambridge, U.K. Thompson, P. (1982). Perceived rate of movement depends on contrast. Vision Research, 22, Thompson, P. & Stone, L. S. (1996). Contrast dependence of perceived flicker rate: conterphase gratings don't behave like drifting gratings. Investigative Ophthalmology and Visal Science, 37, S91. Thompson, P., Stone, L. S. & Swash, S. (1996). Speed estimates from grating patches are not contrast-normalized. Vision Research, 36, Watson, A. B. & Ahmada, A. J. (1983). A look at motion in the freqcncy domain. In Tsotsos, J. K. (Ed.), Motion: perception and representation (pp. 1-1). Association for Compting Machinery: New York. Watson, A. B. & Ahmada, A. J. (1985). Model of hman visalmotion sensing. Jornal of the Optical Society of America A, 2, Watson, A. B. & Trano, K. (1995). The optimal motion stimls. Vision Research, 35, Wilson, H. R., Ferrera, V. P. & Yo, C. (1992). A psychophysically mtivated model for two-dimensinal motion perception. Visal Neroscience, 9, Acknowledgements-We thank or sbjects fr their participation, Rob Stone for programming, Preeti Verghese and Brent Better fr comments on an earlier draft. LS was spported by NASA RTOP Some of this work was presented at the Association for Research in Vision and Ophthalmology conference in Fort Laderdale, April 1996.